Publications

Abstract : Two-dimensional pictorial representation of the rear part of human eye furnish diagnostic information about blood vessels, optic disc, and macula. Among these features vessel analysis at disc provides quantitative diagnostic information about different stages of ophthalmic diseases. The alteration in morphology of retinal blood vessels in the neighbourhood of disc region convey important diagnostic measures related to cure, and assessment of cardiovascular and ophthalmologic diseases (glaucoma, diabetic retinopathy, and arteriosclerosis). Hence, vessel analysis near by disc becomes a prime factor for analysis of ophthalmic diseases. Here, a multi-channel generative adversarial network is used for simultaneous segmentation of retinal vessels and disc. The model simultaneously segments retinal landmarks through a single generative adversarial network (GAN) using adversarial learning process. Multi-scale residual convolutional neural network (MSR-Net) is utilized as generator which is capable of generating two channel segmentation maps (vessels and disc region) separately. In the discriminator section, two branches of convolutional neural network (CNN)-based binary classifiers are used. The segmentation performance is evaluated on two publicly available databases namely CHASE_DB1, HRF databases and DRIVE database. Different quantitative performance measures are conducted to compare the performance of the proposed method with state-of-the-art methods. The projected work achieved an accuracy of 0.9730 for HRF data set, 0.9861 for CHASE_DB1 data set and 0.9816 for DRIVE data set for segmenting blood vessels. Simultaneously, this method achieved an accuracy of 0.9982 for HRF data set, 0.9965 for CHASE_DB1 data set and 0.9968 for DRIVE data set for segmenting disc region. The proposed method can be used for analyzing diagnostic information about ophthalmic diseases like glaucoma and visual field defects cause due to presence of abnormalities in the optic disc region.